No abstract

A legacy system is an operational, large-scale software system that is maintained beyond its first generation of programmers. It typically represents a massive economic investment and is critical to the mission of the organization it serves. As such systems age, they become increasingly complex and brittle, and hence harder to maintain. They also become even more critical to the survival of their organization because the business rules encoded within the system are seldom documented elsewhere.Our research is concerned with developing a suite of tools to aid the maintainers of legacy systems in recovering the knowledge embodied within the system. The activities, known collectively as "program understanding", are essential preludes for several key processes, including maintenance and design recovery for reengineering.In this paper we present three pattern-matching techniques: source code metrics, a dynamic programming algorithm for finding the best alignment between two code fragments, and a statistical matching algorithm between abstract code descriptions represented in an abstract language and actual source code. The methods are applied to detect instances of code cloning in several moderately-sized production systems including tcsh, bash, and CLIPS.The programmer's skill and experience are essential elements of our approach. Selection of particular tools and analysis methods depends on the needs of the particular task to be accomplished. Integration of the tools provides opportunities for synergy, allowing the programmer to select the most appropriate tool for a given task.

The effective synergy of a number of different techniques is the key to the successful development of an eficaent Reverse Engineering environment. Compiler technology, pattern matching techniques, visualization tools, and software repositories play an important role for the identification of procedural, data, and abstmctdata-type related concepts in the source code. This paper describes a number of techniques used for the development of a distributed reverse engineering environment. Design recovery is anvestigated through code-to-code and abstmct-descriptions-to-code pattern matching techniques used to locate code that may implement a particular plan or algorithm. The code-tocode matching uses dynamic progmmming techniques to locate similar code fragments and is targeted for large software systems ( I M L O C ) . Patterns are specified either as source code or as a sequence of abstmct statements written in an concept language developed for this purpose. Markov models are used to compute similarity measures between an abstract description and a code fmgment in terms of the probability that a giuen abstract statement can genemte a given code fragment. The abstmct-description-to-code matcher i s under implementation and early experiments show it is a promising technique. 'This work is funded by

No abstract

The integration of a cache memory into a connectionist language model is proposed in this paper. The model captures long term dependencies of both words and concepts and is particularly useful for Spoken Language Understanding tasks. Experiments conducted on a human-machine telephone dialog corpus are reported, and an increase in performance is observed when features of previous turns are taken into account for predicting the concepts expressed in a user turn. In terms of Concept Error Rate we obtained a statistically significant improvement of 3.2 points over our baseline (10% relative improvement) on the French Media corpus.